A mixed methods investigation of end-of-life surrogate decisions among older adults

There have been qualitative investigations of surrogates’ experience of making these decisions after they have taken place [ 4, 8– 11]. These show that surrogates do consider the patient’s wishes, either by recalling previous conversations or their shared experiences, but they are not the sole focus of their accounts. Other factors can conflict with deciding in accordance with the patient’s wishes, such as their own values or preserving the patient’s life or the family’s well-being. Indeed, surrogate decisions are often biased towards the decision-maker’s own preferences [ 12, 13]. Consequently, the substituted judgment standard cannot always be met.

Parallel to this literature, a strand of experimental research has found that we are more likely to choose the option with the lowest risk of death for others [ 14, 15]. However, for ourselves, we are more likely to choose the option which might increase our chances of dying in order to avoid an illness [ 16] or complications from taking a treatment [ 17]. This has been explained by professional accountability when medical professionals make decisions [ 18]. When the general population decide for a stranger or family member, the reasons are not clear [ 19– 21]. Interestingly, discrepancies have been found between surrogate choices and predictions. Surrogates avoid a risk of death for someone else more than themselves, despite predicting that they have similar preferences [ 22– 24]. This suggests that surrogates override the recipient’s preferences to make a more cautious decision on their behalf which preserves their chances of living.

Tunney and Ziegler’s model [ 2] puts forward that surrogates engage in various forms of perspective taking when making a decision. The importance they give to each perspective depends on the nature of the decision. For end-of-life decisions – highly significant decisions for which the surrogate can be held accountable – the model predicts that surrogates focus on what is required of them. If the best interests standard should be followed, the surrogate will engage in benevolent perspective-taking (i.e. decide based on what the recipient should do). If the substituted judgment standard should be followed, the surrogate will engage in simulated perspective-taking (i.e. predict what the recipient would do and decide accordingly), particularly if they are close to the recipient and know their wishes. However, the surrogate might also engage in an egocentric perspective (i.e. decide based on what the surrogate wants) to preserve their own interest. If the surrogate has incomplete knowledge of the recipient’s preferences, they might rely on a projected perspective (i.e. decide based on what they would do if they were the recipient). Although the model expects a next-of-kin to focus on a simulated perspective, other perspectives might come into play which prevent strict adherence to the substituted judgment standard.

Other theories make predictions regarding when self-other differences occur. The presence of a hot-cold empathy gap between the surrogate and the recipient would lead surrogates to underestimate the intensity of the recipient’s emotional state, such as a patient’s pain [ 25]. The risk-as-feelings hypothesis expects reduced emotional reactions to the prospect of a risk when deciding for someone else [ 26]. Both assume that surrogate decision-making is no different from one’s own decision process – self-other differences are only a reflection of the psychological distance between the surrogate and the recipient, i.e. how close they are to each other [ 27]. Finally, social values theory suggests that social values are the key factor taken into account when making surrogate decisions and self-other differences should arise when taking a risk is socially valued or frowned upon [ 28]. None of these accounts are able to capture the intricacies of making surrogate end-of-life decisions, which is likely to be a much more complex and reflective process. Tunney and Ziegler’s model [ 2], in assuming that surrogates engage in perspective-taking, is more able to support an understanding of how surrogates navigate such complex medical decisions. It can conceptualise the tensions felt by surrogates when needing to abide by the substituted judgment standard and is therefore well suited to analysing surrogate decisions.

So far, quantitative studies have looked at self-other differences in treatment scenarios, whereas qualitative studies have focused on the experiences of surrogates without taking into account the specific clinical content of decisions made. We do not yet know much about end-of-life scenarios other than qualitative reports taken place after the fact. There is therefore scope for research that can bridge the gap between the decisions that surrogates make and the reasons they give for doing so. In the present study, we investigate how older adults make end-of-life decisions for their partners via a decision-making task and a semi-structured interview. By taking a mixed methods approach, we position ourselves within a pragmatist epistemological framework whereby we accept that qualitative and quantitative methodologies can hold conflicting ontological and epistemological assumptions, but put these aside in our analysis to focus on addressing the research question and its real world implications [ 29].

We used an expansion design with mixed methods to extend the scope of inquiry to different inquiry components [ 30], namely the outcome of the decision (quantitative method) and the process of the decision (qualitative method). Our quantitative research question was: are people more willing to accept a life-saving risky treatment for themselves than for their long-term partner? Given previous findings, we expected participants to accept more treatment for their partner, even if that means risking their quality of life. Our qualitative research question was: which perspectives do surrogates take when making end-of-life decisions for their partners? We conducted in-depth semi-structured interviews which we analysed using a thematic analysis. We recruited older adult partners as they are more likely to make these sorts of decisions for each other in the near future. We could then compare participants’ own decisions to the ones their partner made for them to assess whether any inaccuracy is related to failing to take a simulated perspective (i.e. decide according to their partner’s wishes). We integrate results from both methods in our discussion.

We recruited 12 older adult partners (60–80) in long-term relationships from Nottinghamshire, UK. Recruitment methods included the School of Psychology’s community sample, local University of the Third Age (U3A) branches and word of mouth. Participants were either contacted via email or directed to the investigators via email. Participants took part between April and July 2018. All were able to understand and complete the decision-making task and take part in the interview.

Partners were asked to come to the University of Nottingham together. After giving informed consent, they took part in the study in turn, whilst their partner waited in a separate room. Every participant completed the decision-making task first, which was followed by a semi-structured interview. The ethical and medical context in which surrogate decisions are made were not described to participants to avoid biasing their answers. For example, we did not mention the substituted judgment standard at any point. Participants were debriefed together once both had completed the study.

Participants completed three scenarios adapted from the willingness to accept life-sustaining treatment (WALT) instrument [ 31] (see Supplementary File 1). Each scenario depicted a life-threatening situation in which participants are taken to hospital and offered a high-burden treatment course to recover. The probability of the treatment working varied from 90 to 10% in decrements of 10. In each case, participants had to indicate whether they would want the treatment or not. They were told that without the treatment they would not survive. The outcome of the treatment varied: either the treatment works and their current health is restored, or it does not and they die from the illness (death scenario), end up bedbound (functional impairment scenario) or unaware (cognitive impairment scenario). The functional and cognitive impairment scenarios allow us to examine the risk of impaired quality of life participants are prepared to take to for a chance of living. Every participant completed the scenarios once from their perspective (i.e. making decisions for themselves) and once making decisions on behalf of their partner. The order in which participants completed these was counterbalanced.

To investigate self-other differences, we computed the average between the lowest chance of recovery participants accepted and the highest chance of recovery participants refused: the point at which they were indifferent between accepting and refusing treatment. To assess whether surrogate decisions were accurate, we computed the difference between surrogate decisions and the recipient’s decisions by subtracting the latter from the former and removing the sign. This gave us a value representing how far surrogate decisions deviated from the recipient’s decisions. We consider a result to be statistically significant at p < .05. However, as our sample is small, we will also examine effect sizes which can be more meaningful than p-values.

Participants took part in in-depth semi-structured interviews conducted by the first author (see Supplementary File 2 for interview guide). The questions were open-ended and designed for participants to freely speak about their experience and thought processes in the decision-making task. The questions were centred around three topics: recall of surrogates’ thoughts when making decisions for their partner, further exploration and discussion of their reasoning and experience, and how their surrogates decisions compared to their own decisions. Interviews ranged from 15 to 45 min, were audiotaped and transcribed verbatim.

We analysed the interview data using a thematic analysis which allows us to identify and analyse patterns in rich detail.. We were guided by an essentialist/realist epistemological approach which reports experiences, meanings and the reality of participants [ 32]. We interpreted participants’ motivations and experiences in a straightforward manner, assuming a largely unidirectional relationship between their language and the experiences they report. We took a semantic approach whereby themes were identified within the explicit content and meanings of the data, moving from a description to an interpretation of it. Given our interest in understanding the perspectives participants took when making surrogate decisions, we followed a theoretical thematic analysis driven by the forms of perspective-taking laid out in Tunney and Ziegler’s model [ 2]. This meant that, although we did not start out with an a priori coding frame, our analysis was driven by our theoretical interest and provided a more detailed account of a particular aspect of the data, rather than a rich description of the entire data set. We directed our analysis towards the decisions participants made for their partner, rather than the ones they made for themselves, as our research question was focused on the surrogate decision process.

We followed the analytical steps as laid out in Braun and Clarke [ 32], After transcribing the interviews, the first author (EB), who is trained in thematic analysis, worked through the data set to generate codes using NVivo. EB then went through the dataset again to check that all extracts that were pertinent to our codes had been identified, collating codes if necessary. Once the list of codes had been generated, the coding was checked with a researcher independent to the study (WD) who is trained and experienced in a range of qualitative analysis methods. WD was given half of the transcripts ( N = 12). WD coded the transcripts independently, which was then compared to EB’s coding. EB and WD discussed discrepancies and independently revised their coding, after which the kappa agreement score was 0.98. After this process of triangulation, EB sorted codes into potential themes by considering how different codes might combine to form an overarching theme. EB read through the data extracts of each candidate theme to ensure that they formed a coherent pattern. Next, EB considered whether the thematic map formed a coherent representation of the entire data set by reading through it again. Finally, EB named and defined each theme in order to form an accompanying narrative.

We recruited 12 partners ( n = 24) who were all in heterosexual relationships. Participant characteristics can be found in Table 1. The proportion of participants with children and grandchildren was inferred and collated from comments participants made during their interviews. It could not be exactly defined as 92% ( n = 22) mentioned children and 67% ( n = 16) mentioned grandchildren.

Given our small sample size, we took measures to ensure that we could draw meaningful conclusions from our findings. We calculated the effect size we could detect from our sample. With a sample of 18 participants, a large effect size (d > 0.80) can be detected with adequate power (>.80) and an acceptable alpha level (<.05) according to G*Power 3.1 [ 33]. We calculated post hoc power for our statistically significant results. Further, a member of one of the dyads had very specific health preferences, which stood in contrast to the rest of our sample. This was reflected in their own choices and the ones their partner made for them, and spoken about at length during both of their interviews. Given that we had a small sample, we decided to exclude this dyad from our quantitative analyses to preserve statistical validity. This did not have an impact on the direction of the results, as can be seen in our analyses with the full sample in Supplementary File 3.

We identified three themes, each composed of two sub-themes: respecting their partner’s wishes (with sub-themes ‘beliefs’ and ‘process’), overcoming the uncertainty (with sub-themes ‘drawing from past experiences’ and ‘reproducing their own decision-making’) and balancing perspectives (with sub-themes ‘their partner’s best interest’ and ‘thinking about their own interest’) (see Fig. 3).

Participants were more inclined to accept a life-saving treatment for others than for themselves, which is in line with previous research showing that people are more likely to favour a life-saving choice for others. This is concurrent with participants reporting they did not want to lose their partner and felt that they should give them a chance of living. We found discrepancies between surrogates’ choices and those made by the recipient, indicating evidence of surrogate inaccuracy. However, these results alone do not indicate whether participants intended to go against their partner’s wishes: some might have made a best-informed guess from their knowledge of their partner but got it wrong, whereas others could have known what their partner wanted but chose to make a different decision. Indeed, surrogates reported taking a variety of perspectives to inform their decision-making. Nevertheless, the majority of surrogates intended to decide according to their partner’s wishes and held beliefs that aligned with the ethical underpinnings of the substituted judgment standard.

Most participants held the view that they would rather die than end up with a severely compromised quality of life, and showed that they knew their partner did too. Crucially, this indicates that the source of surrogate inaccuracy might not reside in the fact that surrogates misjudged their partners’ preferences concerning the choice outcomes, but rather that they misjudged their risk preferences. Surrogates did find the cut off level difficult to judge for their partner. Although the reports showed that participants overall held similar views regarding quality of life, they held quite different intuitions regarding the percentage risk that would be ‘too risky’. It is likely that they believed their partner would hold the same intuitions and would only majorly adjust their risk preference if they thought their partner’s wishes differed from their own.

Using a mixed methods approach has enabled us to consider both the process and the outcome of the decision in greater depth. The investigation of self-other differences on their own is nowhere near sufficient to understand surrogate decision-making. Drawing from participants’ reports is necessary to address the complex processes at play. Similarly, identifying how participants’ reports match up to the decisions they made allowed for further nuance in understanding their decision process.

Our findings lend support to predictions made by Tunney and Ziegler’s model [ 2] (detailed in the Background section). Indeed, we found that surrogates intended to make a simulated decision for their partner in end-of-life scenarios. However, no matter how well surrogates felt like they knew their partner’s wishes, there was a remaining level of uncertainty for some which they had to overcome. They therefore drew on their own decision-making, which is in line with Tunney and Ziegler’s [2]prediction that surrogates might default to a projected perspective. Moreover, participants considered a benevolent perspective by thinking about whether the treatment they would put their partner through was in their best interest. Finally, participants took an egocentric perspective when thinking about their own wishes for their partner.

It is clear that surrogate decisions can be a lot more complex than suggested by the theories we outlined in the introduction, namely hot-cold empathy gaps [ 25], the risk-as-feelings hypothesis [ 26] and social values theory [ 28]. Even though self-other differences might seem like they can be explained by these accounts, neither are able to capture the complexity of the decision process and the intricacies across surrogates’ experiences. Tunney and Ziegler’s model [ 2] is far more able to highlight these details.

Participants spontaneously indicated a willingness to honour their partner’s wishes, meaning that the fundamentals of the substituted judgement standard are not necessarily misguided and should not be done away with as an ethical framework. However, the problems previously raised about the substituted judgment standard were highlighted in our study. Surrogates found it difficult to ignore other factors, such as what they want for their partner. A few mentioned that if their partner’s present wishes were considered unreasonable or discordant with their past wishes, they would override them. Expecting the surrogate to make a decision they deem unreasonable magnifies the burden placed on them. Participants also conjectured that it would be more difficult to follow in actual fact due to emotional influences, which is reflected in studies of surrogate decisions after the fact [ 4, 8]. The present study therefore shows that, in principle, surrogates would like to follow the substituted judgment standard, but this is not always achievable in practice.

Another problem with the substituted judgment standard is that it is entirely focused on the patient and does not address the burden on the decision-maker. Participants who had not had previous discussions with their partner were in the dark about their partner’s wishes, whereas those who had were more confident and comfortable with their decisions. Encouraging people to have these discussions in the later part of their lives would be a good strategy to ease the process. Indeed, a retired health care professional insisted that these discussions between family members do not happen enough prior to the event. Promoting advance care planning practices could be a way to encourage people to discuss their healthcare preferences with their loved ones. To be most effective, it should consider factors which can help surrogates make decisions in all kinds of scenarios, such as the chance of recovery following treatment as our study shows that partners may not be aware they have different risk preferences.

Our study used hypothetical scenarios. Although some participants were certain they would make the same decisions in real life, it was apparent that others were not. Some mentioned that the fear of death would be more imminent in a real scenario, and although they would like to think that they would stick to their current decisions, they felt that they might be led towards a different direction. Indeed, in reports which took place after the fact, the tensions felt by surrogates which prevented them from honouring the recipient’s wishes were more apparent [ 4, 8, 9].

Our findings might not be easily generalisable to the wider population. It is conceivable that the surrogate decision differs by demographic (education, socioeconomic status, religion etc). Although we did not collect extensive demographic information, it can be inferred that participants were numerically literate given that they were all able to complete the task which involved making judgments about probabilities. Moreover, most participants were open to speaking about end-of-life. The recruitment process clearly stated that the study would involve thinking about severe illnesses and death, which would have discouraged those unwilling to do so. The participants we recruited would presumably be more likely to have these discussions with their partner. Participants who are less numerically literate and open to speaking about end-of-life might have made different decisions. Future research should be extended to different populations to investigate whether their decision process is different to our findings.